// the find
Jazee6/cloudflare-ai-web
Cloudflare AI Platform with one-click deployment
A Next.js chat UI that sits in front of Cloudflare Workers AI (with optional Gemini access via AI Gateway), deployable with one click to Vercel or as a Docker container. It's for someone who wants a private ChatGPT-style front end without paying for OpenAI, using Cloudflare's inference credits instead.
Actually has tests for the boring-but-important parts (auth, request validation, image route, model catalog) instead of just UI components. It also ships ADRs explaining real decisions (why Cloudflare is treated as the authoritative model catalog, why access sessions are stateless) — rare for a project this size and useful if you're deciding whether to fork it. Deploy story is genuinely one-click: Vercel button or a single docker run with two env vars.
Chat history lives only in browser local storage — no server-side persistence, so switching devices or clearing browser data loses everything. Without setting APP_PASSWORD, the deployed instance has an open inference API that anyone can hit and burn through your Cloudflare quota, and that's opt-in rather than the default. Model support is really just Cloudflare Workers AI plus Gemini through AI Gateway, not a general-purpose multi-provider router despite what the name suggests. Primary docs are Chinese-first with a secondary English translation, so env var nuances (e.g. how to scope the CF_WORKERS_AI_TOKEN) may lag in the English version.